Medical electronic imaging devices, such as radionuclide scanners, ultrasound scanners, and computerized axial tomographic devices produce electrical signals which are potentially subject to objective image analysis using inexpensive laboratory computers to extract diagnostic image features. Subjective interpretations of radionuclide liver scans result in error rates which vary widely among physicians. We propose to devise objective, quantitative means to extract significant liver image features. Examples of extractable features include texture of image radioactivity distribution, which may reflect diffuse liver diseases such as cirrhosis, and changes in size of focal defect in radioactivity distribution which may reflect treatment response of cancer deposits in the liver. Several other feature extraction tasks are proposed. Our preliminary work has shown the general feasibility of the approach. We have developed the necessary information acquisition methods, and have devised a composite texture index which currently shows a 0.12 probability of misclassification of independently diagnosed patients. Pattern recognition techniques are used to further optimize the diagnostic value of this parameter, by itself and in combination with other image features and clinical data. The accuracy and reproducibility of the objective methods developed will be compared with subjective interpretations by physicians. BIBLIOGRAPHIC REFERENCES: Jain, V. K., Eikman, E. A., Flower, D. L., Madden, J. A. A Method for Computer Analysis of Texture in Radionuclide Liver Images. Proc. of the San Diego Biomedical Symposium 14: 167-173 (1975). Madden, J. A., Jain, V. K., Mack, G., Eikman, E. A. Development of a Fortran Pattern Recognition System for a Minicomputer. Proc. 6th Symp. on Sharing of computer Programs and Technology in Nuclear Medicine, 378-384 (1976).